How will a16z’s 30 trillion AI Agent dream be realized? The answer lies in the “AI Hunger Games”

Darwinian AI is solving the problem of capital formation and is the driving engine behind Crypto AI innovation.

Written by: 0xJeff

Translated by: AididiaoJP, Foresight News

It has been exactly one year since the wave of AI Agents began in Q4 2024.

At that time, @virtuals_io was the first to propose the concept of "AI Agents Tokenization," meaning the pairing of AI applications/tokens with fairly launched tokens.

In just this short year, the Crypto AI sector has undergone a dramatic transformation: it has driven the open-source movement for general AI, and a wide variety of tools have emerged, making it easy for both developers and beginners to build projects.

Initially, it was just an AI product issuing a token, with a low-valuation fair launch, led by independent developers or small teams. Now, it has developed into a complete Crypto AI ecosystem, with hundreds of outstanding teams building their visions here.

Given the recent hype brought by the x402 narrative, this article will explore the most important questions—where is all this heading? What is the core value of Crypto AI Agents?—by reviewing the current state of the industry, understanding the changes, and analyzing the progress of key players.

If, like me, you are excited about AI and eager to learn, you have probably noticed how fast AI is developing. Every month, something new and cool appears. From basic applications that are "nice to have," like Ghibli-style everything, to AI-generated videos of production quality, and even AI Agents whose productivity surpasses that of ordinary junior programmers.

But in the Crypto space, things are not always like this. When the AI Agent narrative emerged around this time last year, the hottest projects were:

• @truth_terminal became lifelike, interacted with a16z's @pmarca, and received investment.
• @aixbt_agent provided sharp analysis and was also a quirky, native Crypto player on the X platform.
• @virtuals_io, as the "Agent Society," saw its "Agent Tokens" skyrocket by 10-50x at times.
• @dolos_diary was the internet's number one "bully," loved for its sharp wit and humor.
• @luna_virtuals as the first AI idol.

When the narrative started, entertainment was the top theme. But now, we haven't seen AI Agents bring any new forms of entertainment for a long time (which might be a good thing, but the charm and appeal of the early AI era have clearly faded).

Now, the focus is intensely on the verticals where Crypto excels: financial use cases, i.e., making money (and not losing money).

a16z, in its latest "State of Crypto" report, proposed a potential market size of $30 trillion for the agent economy. This may be somewhat unrealistic, as the entire AI market is only expected to reach several trillions of dollars by 2030.

That said, I do believe the entire agent economy could indeed be worth trillions of dollars. As generative AI tools and vertical AI help individuals boost productivity, enterprise adoption increases, and more efficient AI-driven workflows are introduced and implemented within organizations, this market will continue to expand.

The Crypto sector is no exception. But since this industry is extremely focused on making money, its workflows naturally revolve around profit. The following categories stand out in particular:

DeFi: The Most Mature Product-Market Fit in Crypto

• Trading (spot, perpetual contracts, on CEX/DEX)
• Money markets (lending, collateralizing crypto assets)
• Stablecoins (medium of exchange/unit of stable value, composable high-yield DeFi strategies)
• Yield protocols (interest rate markets, points markets, funding rate markets, yield optimizers/vault products)
• RWA/DePIN (tokenizing real-world productive assets, connecting on-chain capital to off-chain demand)

This is the largest potential market, with total value locked exceeding $150 billions and stablecoin market cap over $300 billions. Increasing regulatory clarity and institutional adoption are driving more capital on-chain; the surge in stablecoin adoption is also attracting more enterprises and startups to use crypto rails.

For these reasons, the demand for automation as backend infrastructure and tools, with enterprises/startups as the frontend bringing ordinary users on-chain, will be key to driving the next phase of adoption.

AI agents that can abstract away DeFi complexity, simplify execution, or improve key aspects of DeFi (such as risk management, asset rebalancing, strategy curation, etc.) are likely to capture a significant portion of the massive value flowing into DeFi protocols.

Key Ecosystem Players:

@almanak, @gizatechxyz, @Cod3xOrg, @TheoriqAI, @ZyfAI_

• DeAI is the most mature product-market fit in Crypto AI
• Prediction Markets x AI: The fastest-growing Crypto vertical

If you keep an eye on the ecosystem, you'll notice little change in the DeFi x AI space. That's because cracking DeFi workflows is extremely difficult. You can't just throw AI in and hope for good results; responsible structural design and safeguards must be implemented to prevent serious incidents.

Why am I talking about these, instead of generic "AI Agents"?

The initial AI agent ecosystem was basically Virtuals and the agents built within its ecosystem (perhaps with a few others like CreatorBid), as well as frameworks like ai16z (now called ElizaOS), which made it easy to build "agents" or X bots that could call various tools, plus many other frameworks like Arc and Pippin.

These things are cool and interesting, but they are not the true definition of AI agents. A real agent should be able to understand its environment, its own role and responsibilities, make decisions and take actions proactively, and achieve specific goals with minimal human intervention.

Looking around, over 95% of projects are not like this. They are either just software, a generative AI product, or still in the process of evolving into autonomous AI agents.

I'm not trying to disparage anyone. What I want to emphasize is that we * are still at a very early stage, so much so that most people haven't really figured out what works.

Those who have figured out what works are usually not classified as "AI agents," but rather as AI projects.

Current State of the Ecosystem

The recent hype brought by x402 has stimulated capital rotation and interest in Crypto AI, but the new ecosystem looks very different from before.

1. The Hype Around Frameworks Has Faded

Frameworks used to be very important, helping builders get started quickly and reducing the time spent learning and writing code or designing workflows. Tools like MCP improved agent API calling or provisioning capabilities, ERC-8004 will help establish registries and cement Ethereum as the trust and settlement layer, Google's A2A & AP2 are becoming the frameworks of choice for builders, and AI agent/workflow builders like n8n are attracting many developers and ordinary users.

Because of this, the hype around "frameworks" themselves has cooled, and many projects have shifted in other directions. For example, @arcdotfun has pivoted to workflow builders; @openservai, initially positioned as a "cluster," has also shifted to workflow builders and tools aimed at leveraging agents to create Web3 AI-driven businesses for specific user groups (such as prediction market workflows).

Frameworks are still important, but with the popularity of Web2 AI frameworks and tools, and the adoption of Web3 rails, the hype around Web3 frameworks has subsided.

2. Industry Model Transformation

The fair launchpad model benefits small retail investors but makes it difficult for teams to scale. It is also prone to becoming a hotbed for independent developers to build short-term or purely speculative projects, rather than long-term AI businesses that can last 3-5 years or more.

In this regard, it makes sense for Virtuals to expand through its agent business protocol. As x402 establishes itself as the agent payment channel, building infrastructure for agent trust/reputation scoring and mechanisms for agents to collaborate and pay each other for services is crucial to realizing the agent vision.

However, challenges and core questions remain: "Are there high-quality services people are willing to pay for?"

If most services are useless, why wouldn't people just use Web2 AI services instead of Web3? If that's the case, what's the point of gathering Web3 agents together?

To build a sustainable AI business that generates 7-8 figure revenue, you need capital, highly motivated talent, and time to build the vision, which the fair launch model struggles to provide.

Instead, we are seeing medium to large AI teams become increasingly popular, able to raise seed funding from angels and VCs and enter the market via community rounds.

With their resources (capital, talent, VC backing, etc.), these teams can usually deliver much higher quality products/services, which also tends to make their tokens perform better.

3. Broken Profit Models and Tokenomics

Managing both an AI product and a token requires two completely different skill sets, and it takes careful design to combine the two to accelerate product growth and user acquisition (for example: airdrop tokens to the right users → users convert to paying users → pay to use the product → earn more tokens, which, through revenue sharing, buybacks, governance, etc., bind users to the project's long-term interests → the flywheel keeps spinning).

Easier said than done. Most small AI agent teams allocate 30-80% of their tokenomics, leaving no resources to kickstart any growth flywheel.

Most projects use a SaaS subscription model or charge by usage/points, with an option to pay with tokens for a discount. Many projects use part of the subscription revenue to buy back tokens or burn tokens used to pay for services.

Buying back tokens with subscription revenue is fine, but simply forcing token payments (or only offering discounts) is hard to scale.

Crypto tokens are extremely volatile. Using them as a payment medium is not a good idea (they might go up 20% today, down 30% tomorrow—it's hard to budget).

4. Darwinian AI: A New Path for Capital Formation and Clear Tokenomics

@opentensor (Bittensor) has become the go-to platform for founders to launch ideas, miners to contribute to AI, and investors to back the next potentially disruptive DeAI company.

@flock_io leverages federated learning to set standards for privacy-preserving and domain-specific AI, attracting Web2 enterprises, governments as clients, and trainers (miners) who want to contribute to AI. Like Bittensor, Flock helps companies accomplish cool and meaningful AI work with external top talent.

@BitRobotNetwork, inspired by Bittensor, is taking a similar approach to guide a robot-centric subnet ecosystem.

At the same time, real-world benchmarks/evaluations with real money at stake are emerging (which has also become a form of high-quality entertainment):

• @the_nof1's Alpha Arena pits six cutting-edge AI models against each other in a perpetual contract trading competition with real money ($10,000 each).
• @FractionAI_xyz uses competition among AI agents to improve/continuously fine-tune agents to provide better outputs, signals, profits, and risk management.
• @openservai created OpenArena, where AI models compete in prediction market trading.

Darwinian AI is solving the problem of capital formation and is the driving engine behind Crypto AI innovation.

• The top Bittensor subnet, Chutes, is already the number one inference service provider on OpenRouter, which is the world's most popular unified API gateway for general AI developers.
• The top compute subnets (3-4) have a combined annual recurring revenue of $20-30 millions.
• Prediction-related subnets are starting to earn hundreds of thousands to millions of dollars in annual recurring revenue by monetizing alpha signals and/or using signals for better trading/prediction.

Darwinian competitive AI = capital formation (no VC needed) + innovation accelerator (attracting AI/ML engineers to contribute) = this will be the core force driving the AI agent narrative in 2026.

Note: "Darwinian AI" refers to a decentralized ecosystem that drives AI model development, evaluation, and rewards based on competition and market economics. Its core idea is "survival of the fittest," just like Darwin's theory of natural selection, allowing the best and most useful AI models to win in open competition and receive rewards.

So, for small teams or AI agents, what's exciting right now?

Honestly, there are a few that I find useful, but none that I'm willing to pay for yet.

• Research: Grok covers the X platform, ChatGPT covers general domains.
• In-depth analysis: I mainly read newsletters and Messari reports.
• Quick market outlook: I use @elfa_ai's TG chatbot.
• Prediction market trading ideas: I use @AskBillyBets, @Polysights, and @aion5100's @futuredotfun. (I'm looking forward to @sire_agent's aVault, but it's not public yet.)
• DeFi: I mostly operate myself, sometimes using @almanak and @gizatechxyz, but these are not strictly "AI agents."
• Trading: I use @DefiLlama to swap on EVM, or @JupiterExchange to swap on Solana. I don't do perpetuals (if needed, I use @Cod3xOrg for analysis and execution).

The Crypto space is used to letting users use everything for free, so users prefer free tools. Token-gating or paywalls don't work well, but embedding fees seamlessly into the product does. That's why outcome-based pricing models * are very effective. People don't want to pay $40 a month, but are willing to pay $40 in gas fees for a successful trade.

If you can deliver optimal results (high returns, best trade prices), as long as the results are good enough, no one will mind if you build in a fee.

After trying so many Crypto AI applications or agents, what I've learned is: the best products right now are the ones that can make money, and the best verticals to achieve this are launchpads (and the soon-to-explode prediction markets), i.e., running on-chain "casinos" and accumulating fees from trading.

Outlook

• Real use cases that can achieve mainstream adoption (i.e., ones that ordinary AI developers or users outside the Crypto space will use) will emerge next year, most likely from the DeAI/Darwinian AI ecosystem.
• 2026 will be the year of Crypto AI, with DeFi use cases, DeAI infrastructure, and prediction use cases emerging in large numbers.
• Most small agent teams will gradually disappear or be acquired/merged, or shift to building within the Darwinian AI ecosystem.
• Crypto AI and AI agents as verticals will merge, marking a clearer product direction and vision for Crypto AI.
• Launchpads will remain the core of Crypto Twitter, generating trading volume and fees, but the real industry-advancing innovations will happen where resources (capital, talent, distribution channels, and user adoption) are most concentrated.

What is the true significance of Crypto AI Agents?

For "AI agents," the significance lies in designing a trading experience under the guise of "investing in technology," even though most are just LLM wrappers with a token shell.

In most cases, it provides retail investors the best way to speculate on and profit from early investment in these "AI agent" assets.

As a narrative, Crypto AI agents lay the foundation for the future agent economy, where blockchains will serve as the core infrastructure/rails to make it all possible.